Conditional Growth References for Vietnamese Children: Quantile Regression of Growth Distributions and Phenotypes

Background and objective. The WHO growth reference judges a child's height against the population at one age, without using the child's own earlier growth. This study built and validated a quantile regression based conditional growth reference predicting a child's future height for age distribution from early growth status. Methods. Using deidentified longitudinal annual school children health check data from three major cities in Vietnam, this study linked a baseline height and BMI for age Z score near age 6 to outcome visits near age 12 or 15 in 4,857 children. A linear quantile regression and a pooled quantile generalized additive model, fit by sex, were compared against a linear benchmark and a naive persistence model on a held out test set, using pinball loss, interval coverage, and width against the WHO 90% band. This study also tested adding growth phenotype and checked generalization across the three cities. Results. The pooled model outperformed the discrete quantile regression in every group. Every individualized 90% interval was narrower than the WHO band, by 34.6 to 54.5%. Growth phenotype improved accuracy in most groups. The model generalized well to a held out Hanoi or Haiphong but poorly to a held out Ho Chi Minh City, where coverage fell to between 0.35 and 0.52, though the all-city model stayed well calibrated for Ho Chi Minh City itself. Conclusion. Conditional quantile modeling gives individualized growth predictions markedly narrower than a fixed population reference and gains further from growth phenotype, but needs local validation before use at an unrepresented site.